基于BP算法的二维阵面推进网格划分方法的优化与应用OA北大核心CSTPCD

In computational fluid dynamics,mesh quality and generation efficiency are very important to the accuracy and efficiency of numerical analysis.In this paper,a node-screening algorithm is established by using the normal of advancing front to classify and screen the generated grids on the basis of the BP algorithm of unstructured grid front advancing method.We also construct a neural network structure and reset loss function and activation function in order to achieve the optimization effect.Finally,several two-dimensional unstructured triangular meshes are generated by using the optimized method(taking cylinder heat transfer problems,NACA-0012 airfoil flow problems and a group of single joint crack diffusion problems as examples).Mesh quality and generation time are tested using specific methods.The results are compared with the traditional front advancing method and the combined neural network front advancing method before optimization.The results show that,on the premise that the grid quality is not degraded,especially for the grid division of complex shapes,the optimized method reduces the generation time by about 43％,compared with the traditional front advancing method,and about 25％,compared with the combined neural network front advancing method before optimization,which shows that the grid generation efficiency is significantly improved.